Previous incidents

Resolved by fixing the queue edge case that caused worker deadlocks and stalled operations.

A subset of sandbox runners experienced performance degradation during a period of elevated workload volume. This caused internal runner housekeeping and state management operations to slow down significantly.

Because the degradation was not fully captured by existing runner health checks, the sandbox orchestrator continued routing workloads to runners that were not operating at full capacity. This resulted in cascading performance impacts, including increased latency, timeouts during sandbo...